Building for Everyone: The UX Challenge of Consumer AI

What comes after prompts and why AI needs better interfaces for a billion users.

At F4 Fund, when we've looked at Consumer AI companies, we've been asking: "Won't ChatGPT eventually offer what they are building?"

It's clear that in the recent months (April 2025) ChatGPT's image generation capabilities have advanced so dramatically that many standalone photo generation apps in the App Store now appear obsolete by comparison.

But I want to argue why the dozens and dozens of photo generation apps on the App Store will be having a good future. This argument comes from understanding the preference for a certain user experience.

Prompt UX

Personally, I feel like prompt crafting is great and I like figuring out what I can do with ChatGPT, when it comes to creating content and doing much more, like fixing my writing in so many ways.

But to the ordinary folk, the person who isn’t as savvy or doesn’t get a kick out of this text-typing interface, it can feel off-putting to have to write out instructions just to get things done.

Writing text and not knowing what the result will be is not something that the regular folks are used to, and they will continue disliking ChatGPT.

I consider myself a techie to an extent, but I have yet to run anything locally on my machine, join the Midjourney Discord, or play around with the tooling.

There is a spectrum of preference and delight in user experiences when it comes to consumer tech. Imagine the way Google Search became something people felt was valuable, but more importantly, the expectation came from volition. People chose it as their way to surf the web.

Here's what I think is the list of ChatGPT adoption barriers for non-technical users:

1. Contrast with familiar interfaces

Regular folks are used to menus, buttons, or guided wizards on their computers. For mobile apps, there's even more convenience involved, with simple menu structures and guided processes. Now, compare that to writing prompts, which feels so archaic.

2. Mental model mismatch

The whole essence of prompt writing is that it assumes that the user can adopt the mental model of "programming" the AI. This way of interacting with AI is foreign to most people who are used to consuming rather than directing technology.

3. Feedback loop issues

The frustration of prompt trial-and-error can turn non-technical users away from ChatGPT. Unlike traditional UIs with buttons that lead to predictable outcomes, prompt-based interactions create an uncertainty loop. When the AI's response doesn't match expectations, the user faces a confounding dilemma: how to fix it without clear guidance? Like, you're debugging prompts all of a sudden.

Each adjustment feels like guesswork rather than deliberate correction. The uncertainty loop can be very demotivating: after several unsuccessful attempts, many conclude either the technology is flawed or they lack some special skill, leading to abandonment and low NPS. It's a stark contrast with familiar interfaces like mobile apps, where users can follow a suggested path.

4. Cognitive load

When typing instructions for AI, people need to juggle multiple complex considerations simultaneously: a) they must think about the content they want (what specific information or output they need), b) the structure that will work best (how to organize their request for optimal results), and c) various technical aspects (like specifying formats, setting constraints, or using special terminology).

This mental multitasking is exhausting for non-technical users who just want to accomplish a task. Unlike familiar interfaces where options are visible and decisions are sequential, prompt-writing forces users to frontload all these decisions into a single text entry with no clear guidance.

The average person doesn't want to feel like they're programming or solving a puzzle just to get help from technology. When someone needs to draft an email or create an image, having to consciously consider prompt structure creates friction that feels unnecessarily technical and intimidating.

5. Signaling success

In familiar interfaces, success is immediately apparent: a green checkmark appears, a progress bar completes, or a confirmation message pops up. When you upload a file, you see it appear in your folder. When you send an email, it moves to your sent items. These visual confirmations provide instant validation that you've accomplished your task correctly.

With AI prompts, however, users enter a murky zone of uncertainty. They type their request and receive... something. But is it the best something? Is it what they could have gotten with a better prompt? There's no benchmark, no confirmation that they've mastered the interaction. Even when the result seems adequate, users are left wondering if a different approach might have yielded superior results.

Things that are working in ChatGPT's favour

The interfaces we use to interact with AI will surely change as we move into the future. Here are the obvious ways that will help the multipurpose AIs like the ChatGPTs of the world to access non-technical users and become the first billion user consumer services, the Google or the Facebook this technology revolution.

1. Voice UX

If you tried ChatGPT’s “Santa” voice experience this past Christmas with your kids, you probably felt the magic. Voice latency has shrunk to the point where you can hold an actual conversation with an AI on virtually any topic. It’s fast, fluid, and increasingly natural.

That being said, the quality of the interaction still has room for improvement. What I’ve noticed is that the model often jumps to quick solutions or fixes, even when I’m just looking to chat or explore a thought. The emotional nuance and conversational pacing aren’t quite there yet. But the interface? Already impressive.

2. Memory and a Contextual Relationship with AI

ChatGPT’s new memory capabilities are a game-changer, open up the true possibility of a contextual relationship with the AI. It can remember past conversations, context, preferences, and surface relevant information when needed. This creates a sense of continuity, which is a killer feature for AI to feel more like an assistant or collaborator, rather than a blank slate you re-train every time you interact with it.

3. A Fully Embedded Experience

The biggest unlock? An AI experience embedded seamlessly into your everyday life, especially your phone.

Imagine a smartphone where AI is the default interface for everything. Not an app, not a feature, but the the layer that runs the phone. Just like Google won the search wars by being the default in Safari on iPhone, the race is now on to become the default AI layer for the next generation of devices.

Could OpenAI become that? One that sends your messages, edits your photos, generates documents, crafts your Spotify playlists, and anticipates what you need before you ask?

Final words

It’s an incredibly exciting time to be building startups. Consumer AI is rapidly expanding into mainstream adoption, reaching non-technical users at a scale we’ve never seen before.

That’s the direction we’re heading, and in that world, prompts will feel like training wheels, or like the terminal command line we left behind in the 20th century. The interface will evolve, and the opportunity is here now.